Write ad variations for creative testing

Kite builds every variation from customer evidence, because the words a customer chose are more specific and more believable than anything a copywriter invents. They are also what makes the angles truly different from one another. Two angles drawn from two different things customers said will test two different ideas.

Kite looks for that evidence in several places:

• reviews and testimonials in your team's knowledge base or pasted into the conversation
• comments people left on your past ads
• quotes from sales calls and customer surveys
• your own past ads that performed well, read through your connected ad account
• results your customers have reported

Kite also checks what competitors are running. Its built-in ad research shows a competitor's current ads on Meta, Google, and LinkedIn, including the copy, images, landing pages, and run dates. It can find them from the competitor's website address alone, or search Meta's Ad Library by keyword to see what a whole market is saying. That shows which angles are crowded and gives Kite something to contrast against. Competitors' claims stay theirs.

When Kite can't find any customer evidence, it stops before writing and tells you what it searched and what would help, such as a few reviews pasted into the conversation or approval to research what customers say about you in public. If you then ask it to write from your positioning alone, it will, and it marks the batch so you know no customer evidence backs it.

An angle is a reason to click, and each one should come from a different insight about your customers. Kite picks three to five angles before it writes a single line, drawing on these kinds of insight:

• the problem your customer was living with
• the outcome they got
• proof, such as a result a customer reported
• a comparison with the usual way of doing things or with an alternative
• who the product is built for
• an objection answered
• a contrarian view your customers share

An offer or deadline angle only appears when you have a real offer with a real end date. Kite also considers how much the reader already knows. Someone who has never heard of you needs to recognize their own problem in the first line, while someone you're retargeting after a visit to your pricing page already knows the product and needs a reason to act now.

Here's an example. Say you sell scheduling software to physical therapy clinics. Your reviews keep mentioning three things: patients stopped forgetting appointments once reminders went out by text, front-desk staff stopped spending mornings on the phone, and owners worried that switching systems would mean losing their patient history. Those become three angles: the outcome (fewer missed appointments), the problem (a front desk buried in phone calls), and an objection answered (your patient history comes with you). Each angle tests a different idea about why clinic owners buy.

Once the angles are set, Kite writes variations that change how the idea is expressed while keeping the idea itself the same. It varies the opening, the level of detail, and the form of the sentence, so one variation might be a statement, another a question, and another a direct instruction. Lines that differ only by synonyms test nothing, so Kite doesn't write them.

For the clinic example, two variations of the "front desk buried in phone calls" angle might be:

• "Your front desk spends the morning confirming appointments. Let text reminders do it."
• "How many calls did your front desk make to confirm today's patients?"

Both make the same argument, but one opens with a statement and the other with a question. If both beat the other angles, you've learned that the phone-call problem is what moves clinic owners. If the question beats the statement, you've also learned something about how to open the next round.

Kite writes the variations to fit how each platform uses them:

• Google search ads. Google mixes up to 15 headlines and 4 descriptions in one ad and shows any combination. Kite writes each headline so it makes sense on its own and next to any other, and includes headlines that carry your keyword, state a benefit, and ask for the click.
• Facebook and Instagram ads. Meta shows one ad with several text options and learns which combination works. Kite writes two to five primary texts with different openings and two headlines doing different jobs, so every combination reads well.
• LinkedIn, TikTok, and X ads. The feed cuts text off after a short preview, so Kite puts the hook in the first sentence of every variation.

Every line goes through a checker that counts its characters against the platform's limit and flags anything that runs long, repeats another line in the same ad, or is missing a source. Kite fixes what the checker finds before you see the batch.

Kite can make the visuals for the test too. It creates and edits ad images in your brand's colors and fonts and produces short vertical videos. When you supply the image yourself, Kite writes the words to carry what the picture doesn't already show.

Test three to five angles with a few variations each, and no more than your budget can support. Every variation needs enough people to see it before its results mean anything, so a small budget spread across thirty ads tells you very little about any of them. A smaller test with clearly different angles teaches you more than a large test of near-identical lines.

The shape of the batch follows the platform. On Meta, a test might be four ads, one per angle, each carrying two or three primary texts and two headlines that Meta mixes and learns from. On Google, each responsive search ad carries between 3 and 15 headlines and between 2 and 4 descriptions, and Kite gives Google more choices when your customer evidence supports them. If your budget is tight, tell Kite, and you can agree on three angles instead of five so each one gets enough delivery to judge.

Structure the test so that the message is the only thing that changes. Run the variations to the same audience, with the same budget treatment, sending people to the same landing page. If one angle goes to a different page or a different audience, you won't know whether the angle or the page made the difference.

Kite labels every line in the batch with the angle it belongs to, so results can be read back by angle rather than ad by ad. It also points out the platform settings that can quietly spoil a test:

• Meta's own text features can rewrite your ad text after you upload it. Kite reminds you to turn them off for a test, because otherwise you're measuring Meta's rewrite, not your angle.
• Pinning a headline to a fixed position in a Google search ad narrows what Google can test. Kite only pins a line when brand or legal rules require it, and tells you which one and why.

Before the test starts, decide which result counts as winning, such as cost per lead or cost per purchase. Clicks alone can reward a hook that attracts the wrong people. Kite uses the measure you name. If you don't name one, it says which one it used.

Kite delivers the batch in the conversation, in the Kite web app or Slack. You see the angles with the insight and source behind each one, every line of copy with its source, and the checker's results. The same lines come in a file your team can paste from.

Your team picks the variations it wants, adds the landing page link, and launches the test in the ad platform. A live ad spends your money in public under your company's name, so the person who controls the budget makes that call. If you'd rather upload in bulk, give Kite your campaign and ad group names and the landing page URL, and it prepares a copy of the batch in the layout your import needs. With LinkedIn Ads connected, Kite can also build the ads into a draft campaign in your account, and a team owner or admin launches it from Kite.

A creative test pays off in the round after it. With your Google Ads or Meta Ads account connected, Kite reads each ad's results for the period you choose and sorts the lines by angle. The winning angles get fresh variations. The losing angles are retired. One or two new angles go in to test, so every round still explores something new.

Kite is careful about what counts as a result. Lines that didn't get enough delivery to compare are marked as untested, and Kite states the cut-off it used, so a small sample doesn't kill a good idea. Kite never describes a line as tested when there's no performance data behind it.

Each round is recorded in your team's knowledge base: the period Kite looked at, which angles won, which it retired, and which are now under test. The next batch starts from everything earlier tests taught it, so your creative testing builds on itself rather than starting from a blank page.